Anthropic Tops FLI Summer 2026 AI Safety Index at C+

Nine of the industry's most prominent AI developers just got their report card from the Future of Life Institute, and nobody made honor roll. In the Summer 2026 AI Safety Index, Anthropic landed on top with a C+ (score 2.66), OpenAI and Google DeepMind pulled a C (2.28 and 2.01), Meta a D+, and three labs, xAI, DeepSeek and Mistral, received outright F grades. Z.ai and Alibaba Cloud sat in between at D-.

Seven independent reviewers, including David Krueger at the University of Montreal, Stuart Russell at UC Berkeley, Sharon Li, Tegan Maharaj, Sneha Revanur, Robert Trager and Yi Zeng, scored the companies across 37 indicators in six domains: risk assessment, current harms, safety frameworks, existential safety, governance and accountability, and information sharing. Anthropic led five of the six domains; OpenAI took the top slot on risk assessment on the strength of a broader evaluation suite.

The finding that carries the most weight isn't the leaderboard, it's what the panel says the frontier labs have quietly walked back. Reviewers write that Anthropic, OpenAI, Google DeepMind and Meta have weakened or eliminated earlier pledges to pause development if their systems approached specified danger thresholds, replacing unilateral commitments with competitor-contingent conditions. The panel calls this 'moving the goalposts' and argues it has 'undermined safety frameworks across the board'. Russell's version is blunter: the capabilities race, he says, has become more extreme, and companies are 'planning to release them even if it's demonstrably unsafe to do so.' Krueger calls the lack of progress on credible safety plans 'scandalous', noting that even lab leaders are 'starting to get anxious as they race towards recursive self-improvement'.

Existential safety is where the picture is grimmest. For the second consecutive edition, no company scored above a D in that category. The report also flags that firms which once banned military applications, Anthropic, OpenAI, Google DeepMind and Meta among them, reversed course between 2024 and 2026 and embraced defense partnerships.

The honest caveats are worth naming: this is a self-appointed panel with a particular worldview on catastrophic risk, the evidence window closed on 3 June 2026, and grades depend heavily on public disclosures, which reward transparency as much as underlying substance. What the report doesn't give you is the graded companies' own line-by-line responses, or a comparison against evaluations run by the UK and US AI Safety Institutes on the same models. Still, if you're a policy shop, an enterprise buyer, or a board director choosing a frontier vendor for a sensitive workload, the practical message is that even the best-graded lab clears only a C+, and existential safety is uniformly a D or worse. That is a market where independent evaluation and statutory audit regimes still have room to run.